ImageBench V1 —

192 evaluations across 6 categories

Benchmark V1 verdicts are produced by VLM judges and can contain mistakes. Treat PASS/FAIL labels as machine-assisted assessments, and inspect the images yourself. Learn more about the methodology.

Generation Details

Source-backed model context, size, cost, and request settings for this ImageBench V1 run.

local/supra2-img-100m

Local

Supra2-IMG is a tiny 100M-parameter diffusion transformer trained from scratch in under 10 hours on a single H100 (Apache-2.0). It generates fixed 256x256 images with a Flan-T5-Base text encoder, the SD f8 VAE and an Euler flow-matching sampler with CFG; it is a research-scale model, not a production generator.

Maker: SupraLabs
Family: Supra2
Model Size: 100M
high
Cost: local run; no API price
not_applicable
Run Target: gx10/supra2-img-100m
Effective Request: response_format: b64_json · size: 256x256 · num_inference_steps: 50 · guidance_scale: 3 · seed: 7
Sources: SupraLabs/Supra2-IMG model card